The robustness of animated text CAPTCHAs

Abstract

CAPTCHA is standard security technology that uses AI techniques to tells computer and human apart. The most widely used CAPTCHA are text-based CAPTCHA schemes. The robustness and usability of these CAPTCHAs relies mainly on the segmentation resistance mechanism that provides robustness against individual character recognition attacks. However, many CAPTCHAs have been shown to have critical flaws caused by many exploitable invariants in their design, leaving only a few CAPTCHA schemes resistant to attacks, including ReCAPTCHA and the Wikipedia CAPTCHA. Therefore, new alternative approaches to add motion to the CAPTCHA are used to add another dimension to the character cracking algorithms by animating the distorted characters and the background, which are also supported by tracking resistance mechanisms that prevent the attacks from identifying the main answer through frame-toframe attacks. These technologies are used in many of the new CAPTCHA schemes including the Yahoo CAPTCHA, CAPTCHANIM, KillBot CAPTCHAs, non-standard CAPTCHA and NuCAPTCHA. Our first question: can the animated techniques included in the new CAPTCHA schemes provide the required level of robustness against the attacks? Our examination has shown many of the CAPTCHA schemes that use the animated features can be broken through tracking attacks including the CAPTCHA schemes that uses complicated tracking resistance mechanisms. The second question: can the segmentation resistance mechanism used in the latest standard text-based CAPTCHA schemes still provide the additional required level of resistance against attacks that are not present missed in animated schemes? Our test against the latest version of ReCAPTCHA and the Wikipedia CAPTCHA exposed vulnerability problems against the novel attacks mechanisms that achieved a high success rate against them. The third question: how much space is available to design an animated text-based CAPTCHA scheme that could provide a good balance between security and usability? We designed a new animated text-based CAPTCHA using guidelines we designed based on the results of our attacks on standard and animated text-based CAPTCHAs, and we then tested its security and usability to answer this question. ii In this thesis, we put forward different approaches to examining the robustness of animated text-based CAPTCHA schemes and other standard text-based CAPTCHA schemes against segmentation and tracking attacks. Our attacks included several methodologies that required thinking skills in order to distinguish the animated text from the other animated noises, including the text distorted by highly tracking resistance mechanisms that displayed them partially as animated segments and which looked similar to noises in other CAPTCHA schemes. These attacks also include novel attack mechanisms and other mechanisms that uses a recognition engine supported by attacking methods that exploit the identified invariants to recognise the connected characters at once. Our attacks also provided a guideline for animated text-based CAPTCHAs that could provide resistance to tracking and segmentation attacks which we designed and tested in terms of security and usability, as mentioned before. Our research also contributes towards providing a toolbox for breaking CAPTCHAs in addition to a list of robustness and usability issues in the current CAPTCHA design that can be used to provide a better understanding of how to design a more resistant CAPTCHA scheme.